EP0186987B1

EP0186987B1 - Digital signal processor

Info

Publication number: EP0186987B1
Application number: EP85308943A
Authority: EP
Inventors: Edward Roy Pike; Peter Nicholas Pusey
Original assignee: UK Secretary of State for Defence
Current assignee: UK Secretary of State for Defence
Priority date: 1984-12-17
Filing date: 1985-12-09
Publication date: 1991-06-05
Anticipated expiration: 2005-12-09
Also published as: GB8431837D0; DE3583128D1; EP0186987A2; JPH0636165B2; EP0186987A3; US4809210A; JPS61198368A

Description

This invention concerns a digital signal processor in which a mathematical operation is carried out between non-delayed and delayed data.
Such a processor may perform an auto or cross correlation function, or a structure function calculation.
In a typical digital correlator an input signal is divided into successive sample intervals. Digital numbers representative of the signal during each sample interval are clocked serially through the correlator for correlation. A series of digital numbers is clocked along an M-stage shift register to form a delayed signal. A non-delayed series of digital numbers is applied to one input of M different multipliers. The other input of these multipliers is taken from successive stages of the shift register. Each multiplier thus operates on a different delay. The output of each multiplier is accumulated in one of M different counters providing M different channels. At the end of an experimental run the collective content of the counters represents the correlation function of the input signal or signals. This correlator may be termed a linear correlator because the delay between successive channels increases linearly.
An alternative type of digital signal processor measures the "structure function" of the input signal which for long experiments has an equivalent form to the correlation function. To measure this alternative function a "structurator" is required which operates exactly as a correlator described above except that the multiply circuits are replaced by circuits which subtract their two inputs and then square the results.
The correlator may perform an auto correlation or a cross correlation on data. For an auto correlation the input signal is replicated into two identical signals; one signal is delayed and multiplied by the other non-delayed signal. For a cross correlator a first signal is delayed and multiplied with a second, but non delayed, signal.
Digital signal processing enables highly accurate mathematical operations to be carried out on signals. Due to recent advances in logic speed complicated processing can be carried out in real time. Also advances in the statistical theories of some events have simplified the processing of some functions. One example of this is in laser light scattering experiments, particularly in weak scattering events.
Detailed investigation into the properties of light scattering led to the development of a single clipped digital correlator described in U.K. Patent Specification No. 1,290,336. This correlator allowed the processing of signals representing the arrival of single photons on a sensitive detector. From this a whole range of work has been made possible, for example, laser light scattering where the light scattered by a suspension of particles in a liquid can be processed to give particle diffusion co-efficients.
In the above linear digital correlator a correlation function is accumulated from information obtained in successive sample interval channels. Increasing the number of sample channels allows further information to be obtained but results in increased equipment costs.
One solution to the problem of collecting information from many sample intervals is described in G.B. 2,115,192 A. In this specification the time delay between each channel is geometrically increased. Thus for example using 26 channels with a √2 progression between channels information can be obtained from the equivalent of a delay interval of 8192 in a linear correlator e.g. G.B. 1,290,336. The correlator of G.B. 2,115,192 A relies on correlating a signal, at geometrically increasing delays, to obtain the maximum information for a given number of channels. However information is still available from the non-correlated delays. Such uncollected information becomes more important at low counting rates. G.B. 2,115,192 A also applies the principle of increasing delays to a measurement of the structure function.
The present invention extends the data collected from a correlator or structurator such as taught by G.B. 2,115,192 A, without the huge increases in equipment channels that would be involved if the linear correlator of G.B. 1,290,336 were merely extended.
According to this invention in a multi-channel correlator or structurator both the time delay between channels and sampling times in successive channels are progressively increased.
According to this invention a multi-channel digital signal processor comprises:-
means for receiving series of digital numbers in successive sample time intervals (T);
an accumulator for counting the total number of digital signals received;
a plurality of channels having
a series of delays along which a series of digital numbers is transferred to provide a delayed signal, an arithmetic section for receiving at one input a digital signal and at another input the delayed signal, and
an accumulator for receiving the output of the arithmetic section; and
means for increasing the delay (τ) in successive channels
characterised by
a separate series of delays associated with each channel means for progressively increasing the sample time interval (T_s) in successive channels.
The arithmetic section may be a multiplier, in which case the processor is a correlator. Alternatively the arithmetic section may be a difference squarer circuit, in which case the processor is a structurator.
The delayed signal in each channel may be obtained by clocking a digital signal along a multistage multibit shift register.
The means for progressively increasing the channel sample time (T_s) may be frequency dividers set to divide the sample time T by set amounts to clock the delayed signal. These frequency dividers may have fixed values or be changeable to suit differing experiments.
The increase in T_s between channels m and m-1 may be a constant multiplicative factor i.e.

so that

α may be an integer or may take a non-integral value with α^m-1 rounded to the nearest integer. The delay time τ may be a fixed or variable multiple of T_s e.g.

where β is an integer.
The invention will now be described by way of example only with reference to the accompanying drawings of which:-
Figure 1 is a schematic view of the correlator;
Figure 2 is a graph showing the signal to be correlated.
In the realisation shown in Figure 1, the dilation factor α (equation 1) is 2 and the delay factor β is 3.
The correlation circuitry of Figure 1 receives a signal from a photomultiplier, indicated by numeral 1. The photo-multiplier output is a stream of pulses each representing receipt of a photon in a light flux 2. These pulses have a random distribution of amplitude and charge. They are converted to pulses of equal amplitude and duration by the standardisation circuitry 3.
A sample time clock 4 of basic period t can be programmed to provide sample intervals T = ut where u is an integer varied to suit experimental use. This allows the correlator to sample the standardised data on differing time intervals, changing the effective span and range of the correlator in delay time space.
The correlator has M channels, typical M is 24. The number used depends on the nature of the experiment being investigated, frequently only the first 10 or so channels will be needed.
Each channel comprises a counter 5, a shift register 6, a multi-bit multiplier 7, and an accumulator 8. The shift registers 6_1-M are multibit and of any desired number of stages.
In the example shown the multipliers 7_1-M connect to stages 1 and 4 of the register 6_1-M and feed into the accumulators 8_1-M to form a (stationary) time average
as explained later. Here
is the number of pulses received in a sample interval of duration T_s centred at time t_d and the angle brackets denote the average. In other examples of processor (not shown) the multipliers (7_m) are connected to stage 1 and n (n > 3) of the register to suit experimental conditions.
The counters 5_1-M receive pulses from the standardiser 3. Sample time clock pulses of duration T are sent to each shift register via frequency dividers 9 set for this example, as listed in Table 1.
T_s ^(m) is the sample time used in channel m. It is the fundamental clock sample time T multiplied by the number set by the divider circuit 9_m.
The delay tine τ^(m) is the delay in time between the non-delayed and delayed inputs to the multiplier 7_m.
Accumulator 10 accumulates standardised pulses for the duration of an experiment to form
as explained later. Accumulator 11 collects clock pulses of duration T over an experiment to provide N, the total number of clock pulses.
To conduct an experiment the value of u and hence T, (T = ut), is set in accordance with expected results e.g. previous experiments. Typically the value of n can be set between 1 and 3200 and t can be one microsec. Thus T, in this example can be varied between 1 microsec and 32 millisec. The contents of all the correlator stores, counters 5, accumulators 8, 10, and 11 and registers 6, are set to zero. Light from an experiment, e.g. scattered by a particle suspension, is received by the photomultiplier 1 whose output is fed to the standardiser 3. Pulses from the standardiser 3 enter the counters 5 and are clocked at the appropriate channel sample times. At the end of the sample time T_s ^(m) appropriate to channel m in the contents of counter 5_m are clocked into the first stage of the shift register 6_m the previous contents of the first stage are shifted to the second stage and so on. Also the contents of stages 1 and 4 are multiplied in the multiplier 7_m whose output is accumulated in the accumulators 8_m.
An experimental run will typically last many sample times T_s ^(m). At the end of the experiment channel accumulator 8m contains the value
Here N^(m) is the total number of sample times appropriate to channel m which have occurred in the duration of the experiment. It is related to the total number N of sample times T, obtained from accumulator 11, through the relation

where f^(m) is the divide frequency set in divider 9_m. Also at the end of the experiment, accumulator 10 contains the value corresponding to the total number of pulses n_p accumulated during the experiment. The average number of pulses obtained in one sample time T_s ^(m) , appropriate to channel m, is thus
The theory of digital correlation is known for example from "Photon Correlation and Light Beating Spectroscopy" Eds H. F. Cummins and E. R. Pike, Plenum, New York, 1974, and the theory of the digital structure function is known for example from C. J. Oliver and E. R. Pike, Optical Acta 281, 345-58, 1981.
For most digital correlation experiments it is necessary to obtain the correlation function g ⁽²⁾ [τ^(m)] of the input signals defined by

where
is the number of pulses counted in sample time T_s ^(m) centred at time t_d and the <....> represent a long-time average. As an example, in laser light scattering by a suspension of particles,

where

Γ is the decay rate of the correlation function
D_T is the translational diffusion coefficient of the particles
λ, the light wavelength,
Θ , the scattering angle.
The quantity g⁽²⁾ [τ^(m)] is obtained from the accumulator outputs from the following calculation:
The structurator gives an estimate of the same function by the relation
The correlator described above could be adapted in several ways:-
The divide frequencies set in dividers 9 and the shift register stages connected by multipliers 7 could be programmed to any values to suit experimental needs.
The correlator may be adapted via switches (not shown) to connect as a linear correlator on all channels. This could incorporate extending the shift register in the first channel to have the same number of stages as the number of channels. The multipliers 7_{1 to M} would then connect via the switches to the first stage and to successive stages of this extended shift register 6₁.
In later channels where the sample time T_s ^(m) becomes large, say greater than one millisec, but depending on the processor used, the correlator operations described above could be performed in software by a suitably interfaced computer (rather than using the hardware described above).
To perform cross correlation of two different signals, the input of multipliers 7 connected to stage 1 of the shift register 6 in Figure 1 could instead be fed by additional counters, similar to 5, fed in turn by a standardiser, similar to 3, operating on the second signal.
The growing correlation function could be displayed on an oscilloscope after each sample interval T_s ^(m) in the unnormalised form
A restriction on the above correlator arises from so-called temporal integration. For a correlation function of the form of equation 8 the measured correlation function g_m ⁽¹⁾ (τ), obtained using sample time T_s , is related to the ideal function exp(-Γτ) by

(see e.g. Cummins and Pike, references above).
If T_s is the same for all τ (as in a linear correlator) the last factor (sinh Γ T_s )/Γ T_s is independent of τ and therefore constitutes an unimportant multiplicative factor. However when T_s is a function of τ, as in equation 3, the situation is more complicated. Substitution of (3) in (10) followed by expansion of the second factor in (10) gives
For β >> 1 the "correction terms" in the square brackets are relatively unimportant. For example when Γ τ = 1 and β = 3,

and temporal integration effects can be neglected. The same result holds for the structure function.

Claims

A digital signal processor comprising
means (1) for receiving series (2) of digital numbers in successive sample time intervals (T);
an accumulator(10) for counting the total number of digital signals received;
a plurality of channels (CHANNEL 1 ...CHANNEL M) having a series of delays (6) along which a series of digital numbers is transferred to provide a delayed signal, an arithmetic section (7) for receiving at one input a digital signal and at another input the delayed signal, and
an accumulator (8) for receiving the output of the arithmetic section (7); and
means (5) for increasing the delay (τ) in successive channels. characterised by
a separate series of delays (6) associated with each channel, means (9) for progressively increasing the sample time interval (T_s) in successive channels.
The processor of claim 1 wherein the means (9) for progressively increasing the channel sample time (T_s) are frequency dividers arranged to divide the sample time (T) by predetermined amounts.
The processor of claim 2 wherein the frequency dividers (9) can be adjusted to change some channel sample times (T_s).
The processor of claim 1 wherein the delay time (τ^(m)) in all channels has a fixed relationship to the channel sample time, (T_s ^(m)) where m is an integer equal to a channel number.
The processor of claim 4 wherein the increase (α) in sample time (T_s ) between adjacent channels is equal to two and the delay time (τ^(m)) in each channel is equal to three (β = 3) multiplied by the associated channel sample time (T_s ^(m)), where m is an integer equal to a channel number.
The processor of claim 1 wherein the delay time (τ^(m)) in each channel (m) can be varied.
The processor of claim 1 wherein the arithmetic section (7) is a multiplier.
The processor of claim 1 wherein the arithmetic section (7) is a difference squarer.
The processor of claim 1 and further comprising switches for switching the inputs to the arithmetic sections (7) whereby the processor is constructed as a linear processor.